From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pj1-x1030.google.com (mail-pj1-x1030.google.com [IPv6:2607:f8b0:4864:20::1030]) by sourceware.org (Postfix) with ESMTPS id E2BFA385801B for ; Thu, 24 Mar 2022 19:02:41 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org E2BFA385801B Received: by mail-pj1-x1030.google.com with SMTP id g9-20020a17090ace8900b001c7cce3c0aeso1441935pju.2 for ; Thu, 24 Mar 2022 12:02:41 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=wO8VekwRClMRyaas4T7CB4gj4JwFq0vntpzGMbR4ZgY=; b=M9mRsYBnAHt/R7uztpojPzy7uFll2Ieh6vCCmlylfe+fcaJET17WWf1S5yY/INB+5S Qmqaqhj3JqiLVYTG+MYew1vMNZWRrFj4MHsrB7wDLs+KiTcfr5YzVyawYee4lXAgXoAa XgIG5vgShFRgNNhqa/4O76zf9yoHP3SMsaqb5fPwvOIF1HKfL95utgO0cA8lregHRaKv OTsCAryB3COW0g8z6ZFo2dqJKq9DldQGYA6FbjGL69zpxRLaM3+wOm5Gdh3DQKJI8YTo 7zzH+gLdq+xklpKsy3quCbffKaUHGqSuAVhIZQiRRYOMXZdxQqIi+tdVkxeurj5D5/BD dOXg== X-Gm-Message-State: AOAM532wkvbBTwmD8hfH06S61bthgvfmWQnJd6hn474h0zLmq20QV788 kN7L54/7I3+gUG1VuoK7CB+Wb4CghAkhw6y2oUPME7b8XNM= X-Google-Smtp-Source: ABdhPJzIRr4+pbuo9+VQ/uGkJTW7Kd27WgxgG3sbI+I4kaICKDyouWuvjIjGymvaSHA8tzgyan/eVaCUt4oC9UvYQOU= X-Received: by 2002:a17:90b:4b8b:b0:1c6:452c:f76f with SMTP id lr11-20020a17090b4b8b00b001c6452cf76fmr20219351pjb.28.1648148560869; Thu, 24 Mar 2022 12:02:40 -0700 (PDT) MIME-Version: 1.0 References: <20220323215734.3927131-1-goldstein.w.n@gmail.com> <20220323215734.3927131-17-goldstein.w.n@gmail.com> In-Reply-To: <20220323215734.3927131-17-goldstein.w.n@gmail.com> From: "H.J. Lu" Date: Thu, 24 Mar 2022 12:02:05 -0700 Message-ID: Subject: Re: [PATCH v1 17/23] x86: Optimize str{n}casecmp TOLOWER logic in strcmp.S To: Noah Goldstein Cc: GNU C Library , "Carlos O'Donell" Content-Type: text/plain; charset="UTF-8" X-Spam-Status: No, score=-3025.8 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 24 Mar 2022 19:02:43 -0000 On Wed, Mar 23, 2022 at 3:01 PM Noah Goldstein wrote: > > Slightly faster method of doing TOLOWER that saves an > instruction. > > Also replace the hard coded 5-byte no with .p2align 4. On builds with > CET enabled this misaligned entry to strcasecmp. > > geometric_mean(N=40) of all benchmarks New / Original: .894 > > All string/memory tests pass. > --- > Geomtric Mean N=40 runs; All functions page aligned > length, align1, align2, max_char, New Time / Old Time > 1, 1, 1, 127, 0.903 > 2, 2, 2, 127, 0.905 > 3, 3, 3, 127, 0.877 > 4, 4, 4, 127, 0.888 > 5, 5, 5, 127, 0.901 > 6, 6, 6, 127, 0.954 > 7, 7, 7, 127, 0.932 > 8, 0, 0, 127, 0.918 > 9, 1, 1, 127, 0.914 > 10, 2, 2, 127, 0.877 > 11, 3, 3, 127, 0.909 > 12, 4, 4, 127, 0.876 > 13, 5, 5, 127, 0.886 > 14, 6, 6, 127, 0.914 > 15, 7, 7, 127, 0.939 > 4, 0, 0, 127, 0.963 > 4, 0, 0, 254, 0.943 > 8, 0, 0, 254, 0.927 > 16, 0, 0, 127, 0.876 > 16, 0, 0, 254, 0.865 > 32, 0, 0, 127, 0.865 > 32, 0, 0, 254, 0.862 > 64, 0, 0, 127, 0.863 > 64, 0, 0, 254, 0.896 > 128, 0, 0, 127, 0.885 > 128, 0, 0, 254, 0.882 > 256, 0, 0, 127, 0.87 > 256, 0, 0, 254, 0.869 > 512, 0, 0, 127, 0.832 > 512, 0, 0, 254, 0.848 > 1024, 0, 0, 127, 0.835 > 1024, 0, 0, 254, 0.843 > 16, 1, 2, 127, 0.914 > 16, 2, 1, 254, 0.949 > 32, 2, 4, 127, 0.955 > 32, 4, 2, 254, 1.004 > 64, 3, 6, 127, 0.844 > 64, 6, 3, 254, 0.905 > 128, 4, 0, 127, 0.889 > 128, 0, 4, 254, 0.845 > 256, 5, 2, 127, 0.929 > 256, 2, 5, 254, 0.907 > 512, 6, 4, 127, 0.837 > 512, 4, 6, 254, 0.862 > 1024, 7, 6, 127, 0.895 > 1024, 6, 7, 254, 0.89 > > sysdeps/x86_64/strcmp.S | 64 +++++++++++++++++++---------------------- > 1 file changed, 29 insertions(+), 35 deletions(-) > > diff --git a/sysdeps/x86_64/strcmp.S b/sysdeps/x86_64/strcmp.S > index e2ab59c555..99d8b36f1d 100644 > --- a/sysdeps/x86_64/strcmp.S > +++ b/sysdeps/x86_64/strcmp.S > @@ -75,9 +75,8 @@ ENTRY2 (__strcasecmp) > movq __libc_tsd_LOCALE@gottpoff(%rip),%rax > mov %fs:(%rax),%RDX_LP > > - // XXX 5 byte should be before the function > - /* 5-byte NOP. */ > - .byte 0x0f,0x1f,0x44,0x00,0x00 > + /* Either 1 or 5 bytes (dependeing if CET is enabled). */ > + .p2align 4 > END2 (__strcasecmp) > # ifndef NO_NOLOCALE_ALIAS > weak_alias (__strcasecmp, strcasecmp) > @@ -94,9 +93,8 @@ ENTRY2 (__strncasecmp) > movq __libc_tsd_LOCALE@gottpoff(%rip),%rax > mov %fs:(%rax),%RCX_LP > > - // XXX 5 byte should be before the function > - /* 5-byte NOP. */ > - .byte 0x0f,0x1f,0x44,0x00,0x00 > + /* Either 1 or 5 bytes (dependeing if CET is enabled). */ > + .p2align 4 > END2 (__strncasecmp) > # ifndef NO_NOLOCALE_ALIAS > weak_alias (__strncasecmp, strncasecmp) > @@ -146,22 +144,22 @@ ENTRY (STRCMP) > #if defined USE_AS_STRCASECMP_L || defined USE_AS_STRNCASECMP_L > .section .rodata.cst16,"aM",@progbits,16 > .align 16 > -.Lbelowupper: > - .quad 0x4040404040404040 > - .quad 0x4040404040404040 > -.Ltopupper: > - .quad 0x5b5b5b5b5b5b5b5b > - .quad 0x5b5b5b5b5b5b5b5b > -.Ltouppermask: > +.Llcase_min: > + .quad 0x3f3f3f3f3f3f3f3f > + .quad 0x3f3f3f3f3f3f3f3f > +.Llcase_max: > + .quad 0x9999999999999999 > + .quad 0x9999999999999999 > +.Lcase_add: > .quad 0x2020202020202020 > .quad 0x2020202020202020 > .previous > - movdqa .Lbelowupper(%rip), %xmm5 > -# define UCLOW_reg %xmm5 > - movdqa .Ltopupper(%rip), %xmm6 > -# define UCHIGH_reg %xmm6 > - movdqa .Ltouppermask(%rip), %xmm7 > -# define LCQWORD_reg %xmm7 > + movdqa .Llcase_min(%rip), %xmm5 > +# define LCASE_MIN_reg %xmm5 > + movdqa .Llcase_max(%rip), %xmm6 > +# define LCASE_MAX_reg %xmm6 > + movdqa .Lcase_add(%rip), %xmm7 > +# define CASE_ADD_reg %xmm7 > #endif > cmp $0x30, %ecx > ja LABEL(crosscache) /* rsi: 16-byte load will cross cache line */ > @@ -172,22 +170,18 @@ ENTRY (STRCMP) > movhpd 8(%rdi), %xmm1 > movhpd 8(%rsi), %xmm2 > #if defined USE_AS_STRCASECMP_L || defined USE_AS_STRNCASECMP_L > -# define TOLOWER(reg1, reg2) \ > - movdqa reg1, %xmm8; \ > - movdqa UCHIGH_reg, %xmm9; \ > - movdqa reg2, %xmm10; \ > - movdqa UCHIGH_reg, %xmm11; \ > - pcmpgtb UCLOW_reg, %xmm8; \ > - pcmpgtb reg1, %xmm9; \ > - pcmpgtb UCLOW_reg, %xmm10; \ > - pcmpgtb reg2, %xmm11; \ > - pand %xmm9, %xmm8; \ > - pand %xmm11, %xmm10; \ > - pand LCQWORD_reg, %xmm8; \ > - pand LCQWORD_reg, %xmm10; \ > - por %xmm8, reg1; \ > - por %xmm10, reg2 > - TOLOWER (%xmm1, %xmm2) > +# define TOLOWER(reg1, reg2) \ > + movdqa LCASE_MIN_reg, %xmm8; \ > + movdqa LCASE_MIN_reg, %xmm9; \ > + paddb reg1, %xmm8; \ > + paddb reg2, %xmm9; \ > + pcmpgtb LCASE_MAX_reg, %xmm8; \ > + pcmpgtb LCASE_MAX_reg, %xmm9; \ > + pandn CASE_ADD_reg, %xmm8; \ > + pandn CASE_ADD_reg, %xmm9; \ > + paddb %xmm8, reg1; \ > + paddb %xmm9, reg2 > + TOLOWER (%xmm1, %xmm2) > #else > # define TOLOWER(reg1, reg2) > #endif > -- > 2.25.1 > LGTM. Reviewed-by: H.J. Lu Thanks. -- H.J.